All seems serene at the legendary bell labs headquarters in murray hill, n.j. broad green lawns highlight copper roofs aging into an eye-pleasing aqua-green. A beautiful Japanese-style garden graces an interior courtyard.
But behind this tranquillity lies a poorly understood odyssey of upheaval, transformation-and renaissance. The lab’s glorious history-eight Nobel laureates, some 35,000 patents and a tsunami of world-changing inventions from the transistor to information theory-once led many to consider it a national asset. Almost as well documented is the period of “decline,” spurred by a much-lamented and highly criticized 1990s makeover that has seen the lab scale back fundamental science and emphasize applied projects and meeting business objectives.
What’s missing from the picture, though, is an account of Bell Labs’ remarkable resurgence. Changes rocked the lab to its soul over the decade’s first half. But now, on the verge of the millennium-and its 75th anniversary-the venerable establishment has reclaimed its place at the forefront of industrial research. Today’s Bell Labs is hungrier, faster on its feet, and smarter about business than at any time since the Cold War began, playing a vital role in the success of its upstart parent, Lucent Technologies.
What’s more, basic research has not disappeared, as the critics claim. Scores of scientists continue to pursue dreams that may not pay off for decades, if ever-be it wiring up slug brains to find clues to biological data processing or mapping the universe’s dark matter. The critics are right about one thing-pure science doesn’t hold the place it once did. And that brings its own kind of loss. Still, by creating novel ways to balance business realities with far-off explorations, Bell Labs may be trailblazing a new era in corporate research.
Lucent technologies, people here like to proclaim, is the best thing that’s happened to Bell Labs in recent memory. It may seem less than earthshaking to outsiders, but chairman Henry B. Schacht’s decision to place his headquarters inside the lab and feature the R&D arm in the company slogan provided a ringing endorsement unheard of in the old AT&T days. Chatting in his expansive office, current executive vice president of research Arun Netravali reflects that pride by wearing a polo shirt emblazoned with the message: “Lucent Technologies. Bell Labs Innovations.”
Netravali heads the resurgent team. As the hand-picked successor to the lab’s former leader, Nobelist Arno Penzias, Netravali assumed daily control of research with Lucent’s 1996 formation-well before Penzias retired this spring from the senior scientist’s position. But the native of India has been at the lab since 1972. As a Bell Labs engineer and computer scientist, he pioneered digital image and video compression technology-work which last year earned him the prestigious Computers & Communications prize awarded by NEC Corp.
The almost palpable optimism coursing through Lucent is a far cry from the situation only a few years ago, when the lab was thrown for a loop by fast-shifting global competition. Netravali now sees incredible opportunity in the mayhem. Smaller companies and start-ups may move faster and excel in narrow areas, he notes. But Bell Labs’ strength lies in the ability to make sense of and shape the bigger picture-by assimilating technologies from inside and outside its confines, and fitting them into systems.
To fulfill this promise, Netravali maintains, the need for speed is paramount-in evaluating projects, pursuing research advances, creating novel products and adopting outside technologies. Another critical focus is “cannibalization”-the drive to make Lucent’s own products obsolete. For instance, because of the Internet, tomorrow’s data and voice communications will be very different from today’s, a potential mega-disruption to traditional business lines. Research has to be ready with solutions. “The key is how can you become an attacker of yourself-almost like another company might do to you,” explains Netravali. “Let us become better at doing this than some outside company because it’s going to happen anyway.”
These goals could not be adequately met under the old research model-a model that was itself poorly understood. Contrary to popular perception, research has always been a small part of Bell Labs: Of a total workforce of approximately 24,000, only about 1,300 labor on the R’ side of research and development. However, this relatively tiny endeavor has long served as a fountainhead of science and technology. And for decades following World War II, the emphasis fell on being the first or best: publishing papers, setting transmission records, building the most powerful laser diode.
Bell Labs could afford this Ivory Tower modus operandi largely because AT&T was a regulated monopoly allowed to fold a “research tax” into every phone call and sale. But the Bell System’s 1984 court-ordered breakup into seven regional operating companies, AT&T’s dramatic decentralization five years later into a series of business units, and then trivestiture, which saw roughly one-fourth of its researchers assigned to the new AT&T, forced a dramatic change in that outlook.
Starting with Penzias and continuing under Netravali, research has moved to reflect the company’s new place in a fiercely competitive world. Software studies in object-oriented programming, speech recognition, networking and other fields have been beefed up at the expense of robotics and hard-core physics pursuits such as superconductivity that seem unlikely to have an effect on business. Today the labs are divided roughly 50-50 between the physical sciences and software and networking, a more realistic division than the previous 80-20 split. Meanwhile, in addition to maintaining high standards of excellence, managers have been handed responsibility for meeting the company’s technological needs in their particular areas.
Market awareness is central to the new Bell Labs. Scientists and business colleagues interact more regularly with customers and know a lot more about how clients operate than they used to. Since the early 1990s, about half the lab’s researchers have worked with business unit colleagues on specific “joint projects.” Either side can propose such an endeavor-for creating a new switch, a networking technology, or whatever-that is jointly staffed and developed by research and the particular unit. Up to 50 workers staff each project, though most are much smaller. Specific milestones and timetables are created, and researchers sometimes transfer temporarily to the business unit to help launch the products. There’s even a special “breakthrough” category for innovations with strong potential to dramatically reduce costs, improve functionality or create new markets. Typical breakthrough projects get three times a joint project’s staffing and seek to slash in half the normal three-year time to market. Such targeted research strategies have produced a slew of lucrative products. Standout innovations run from numerous fiber optics advances to a low-power digital signal processor (DSP) chip to various internet protocol switches designed to route data with unprecedented speed and quality.
Since its inception Lucent has also operated a New Ventures group that helps spin off inventions outside its core areas of focus. “If we do our research work right we’re going to create lots of pleasant surprises that are technologically exciting and that make more business sense to commercialize outside of our normal business interests,” explains Mel Cohen, vice president for research effectiveness. Under its old style, such products might well have withered on a lab bench. But as of mid-1998, the group had funded nine start-ups based on Bell Labs innovations.
Despite Lucent’s soaring success-its stock has risen 430 percent since the initial public offering in April 1996-research managers profess wariness about going too far to the applied side. The great challenge, notes Bill Brinkman, vice president of physical sciences and engineering research, lies in becoming better attuned to corporate needs but not to “overdo it so badly that you have no science.”
It’s true that “basic” science studies are less numerous than they were in the past and have been scaled back in scope to match more closely areas of core competency-such as lasers, optical communications and materials research. However, the lab is still a place where people from different disciplines mingle in the halls and share ideas through seminars, forums and lectures. And it still harbors an enviable program of over-the-horizon pursuits.
A study of high-impact research papers by Philadelphia-based ScienceWatch showed that in the physical sciences Bell Labs led the world from 1990 to 1997 with nearly 19,000 citations, easily outpacing the 13,020 of runner-up IBM, as well as the world’s top academic institutions. “The science there is in the first rate,” says Tomihiro Hashizume, a specialist in atomic scale structures who worked at Bell Labs before joining Hitachi’s Advanced Research Laboratory in Hatoyama, Japan. The Japanese firm’s impressive 13-year-old institution is dedicated largely to basic science. However, says Hashizume, “I think we have to be a little bit smarter to be Bell Labs.”
Lucent supports scientific studies for several reasons in addition to gaining a direct competitive edge. One is to create a climate of discovery that attracts top scientists who raise research standards and provide bridges to critical university investigations. Basic research can also act as a broad-based insurance policy, since targeted work naturally focuses on areas that are visibly important-and the future will always hold surprises.
Research is aligned into three divisions that cover a gamut of hardware and software relating to communications: Communications Sciences, Computing and Mathematical Sciences and Brinkman’s Physical Sciences and Engineering. All three sustain well-chosen fundamental work. However, when it comes to the lab’s hallmark studies in areas such as solid-state physics, most long-range fundamental investigations take place inside the Physical Research Laboratory run by Cherry A. Murray, part of the Physical Sciences and Engineering division.
Staffed by about 140 researchers, the lab’s activities span physics, materials science, chemistry, computer science, biophysics and astrophysics. Nearly half the efforts look more than 20 years down the road-with virtually all the rest spanning 5-to-10-year horizons. The hope is that all will ultimately bear fruit. In the meantime, it’s expected in the new climate that researchers should be ready, willing and able to bring their expertise to bear on more pressing problems that might arise. Even within this framework, however, there is a striking variety in how closely related the research is to business objectives-as three examples show.
Dancing on the Head of a Pin
From the beginning, the lab’s work on micro-electro-mechanical systems (MEMS) was set up to have both near-term and long-range benefits for Lucent. The goal of this research is to improve communications systems by building miniature machines-microphones, mirrors and more-that are riddled with moving parts but so small that hundreds fit on a pinhead.
The field has exploded in recent years. Because MEMS devices can be fabricated like an integrated circuit on “last generation” equipment, they can conceivably be made for pennies-and thereby become ubiquitous. MEMS sensors already control automotive air bags, and futurists picture these micromachines driving button-sized cell phones that fit on a lapel, or buildings that sense stress changes caused by an earthquake and adjust their structure accordingly. Lucent won’t be making air bag sensors or smart steel. However, explains David J. Bishop, who heads the Microstructure Physics Research Department, “silicon micromechanics has a huge possibility for impacting lots of technologies we care about- particularly optics, acoustics and wireless.”
An early payoff could lie in MEMS-based residential communications systems. The volume of data that can be quickly passed in and out of homes keeps running up against the severe limitations of traditional twisted copper telephone lines. Several schemes have surfaced to ease this problem. Some cable companies, for instance, offer Internet connections over the broadband lines that bring in television pictures. But such alternatives have capacity and reliability limitations, says Bishop. So the ultimate goal is fiber optics, “future-proof” because it offers near-infinite bandwidth with a minimum of maintenance.
Due to copper wire’s limited capacity, separate telephone lines now have to be run from the central phone company office to each home. The same strategy with fiber optics would be prohibitively expensive.
However, since one fiber optic line can handle thousands of phone and data transmissions simultaneously, it might be possible to run a single line to a neighborhood node, then string shorter lines to individual houses-making fiber optics an affordable alternative to copper wires.
Yet there’s still a hitch. Signals are transmitted along fiber optics lines by lasers-power-hungry devices too expensive to provide to every household. Bishop likens the problem to that facing hypothetical explorers on adjacent mountain tops. They communicate by flicking flashlights on and off. After all, that’s basically how optical communications works-only using lasers instead of flashlights. But suppose flashlights are so expensive only one explorer can afford his own. Two-way communications could still be maintained if the flashlight owner leaves his light on-allowing his counterpart to wield a mirror and reflect rays back to the other mountain in a recognizable pattern.
That’s where MEMS comes in. Data would stream into homes in the usual way. But micro-mirrors invented by Jim Walker and Keith Goossen would reflect light back to the central station, simulating lasers in every household for a fraction of the price. Bell Labs has already built mechanical mirrors that can handle in excess of 10 megabits of data per second, nearly 200 times the capacity of today’s 56-kilobit-per-second high-speed modems. Says Bishop, “It is our hope that there will be some limited field trials in the next year.”
It was easy to envision from the outset how the MEMS research related to Lucent’s business goals. But other Physical Research Lab work has a more tangential relation to the bottom line and may take many years to pay off. Take Alan Gelperin, proprietor of the Slug Emporium, a bank of refrigerators crammed with the slithery beings. A 17-year lab veteran, Gelperin is a computational neurobiologist and neuroethologist, meaning he studies the algorithms nerve cells use to produce behavior. He concentrates on slugs-snails without shells-because the creatures possess an intriguing ability to rapidly and reliably learn about odors, and because this “learning” continues even after their brains have been removed from the body for experimentation.
Gelperin works primarily with Limax maximus, the spotted garden slug. The key to devising models that can be simulated in software or even wired into a machine lies in physiological experiments designed to get at how slugs store and access their odor memories, then take action based on their experience with certain scents. In collaboration with colleague Winfried Denk, Gelperin studies dyed slug neurons through two-photon scanning, a microscopy technique that allows him an unprecedented view of the activity inside the processes of single nerve cells.
Similarly, by applying dyes that change their fluorescence if the voltage across the cell membrane changes, he and researcher David Tank, head of the Biological Computation Research Department, have detected electrical waves and oscillations that originate at one end of the odor-analyzer circuit called the procerebral lobe and propagate along it-starting over again as the previous signal dies out. One hypothesis is that the wave acts as a kind of time stamp for storing data. That is, with the detection of an odor and an associated stimulus-a shock, for instance-the memory of that odor is stored in a specific band of cells that run perpendicular to the wave. “Where the wave is determines where the memory storage is going to happen,” Gelperin suggests. The next time the slug is exposed to the odor, it accesses the cells at the same point along the wave-and orders an appropriate response, like gliding away from an odor previously paired with shock. Many experiments remain to be performed before this hypothesis can be confirmed-and possibly incorporated into tomorrow’s neural networks.
But long-range studies are not the only thing Gelperin does. Working with AT&T’s NCR unit before it spun off as a separate company under trivestiture, he used his expertise in neural networks to develop an electronic nose for automated checkout machines. Electronic checkers have little trouble reading bar codes, but they run into real trouble trying to tell a banana from an orange. Gelperin worked with Bell Labs researcher Sebastian Seung, a neural network and machine learning theorist, to create a system that emits a vacuum pulse to pull odors over special sensors that can tell broccoli from lettuce. Last November, Gelperin received a patent on the device.
Gelperin delights in being able to apply his knowledge of neurobiology to solve real-world problems. But he acknowledges that not everybody at the labs has accepted the need to apply their scientific findings. “Some folks just didn’t want to think that way,” he says. “They had their pure science, and pure was with a capital P.’ And they just didn’t want to be bothered.”
90 Percent of the Universe
If gelperin’s research is a fruitful mixture of the basic and applied, Tony Tyson’s seems, at first blush, to be purely fundamental. Tyson is one of the world’s pre-eminent astrophysicists. When his name comes up, Cherry Murray deadpans: “He’s discovered 90 percent of the universe-what can you say?”
Her statement is only somewhat glib, since what the Bell Labs researcher has done is find a way to image cosmic dark matter, the invisible “missing mass” thought to make up some 90 percent of the universe’s total mass. Tyson has made a start on filling in the details. But, he figures, “at the rate we’re currently going it will take me another 50 years.”
The idea that invisible dark matter exists has been around since the 1930s. But the theory attracted only a fringe following until the late 1970s, when modern techniques proved that the visible universe doesn’t contain nearly enough mass to explain the movements of galactic gas and dust-a sure indication something else is out there exerting a strong gravitational effect. Early theories tapped neutrinos for the missing mass, but these particles have since been ruled out as major players. Tyson’s bet is for a combination of unfamiliar objects and events, including weakly interacting massive particles, or WIMPs, magnetic entities called axions, cosmic strings and breakdowns in the uniformity of the space-time continuum.
The 29-year veteran of Bell research has been hunting cosmic dark matter since 1977. “I’m a prospector,” Tyson says. “I should have a donkey, a hat, a canteen and a pickax.” His work makes use of what are called gravitational lenses to map this invisible dark matter. Any mass exerts a gravitational pull that bends or deflects the light from something behind it with respect to an observer. It’s a very imperfect lens-like looking through a Coke bottle. So, if something lies between the Earth and some distant galaxy, for example, astronomers equipped with the right camera sensitivity and processing software will detect multiple images of that galaxy. The distribution of those images makes it possible to figure out how much mass is out there affecting the light.
Dark matter often congregates around visible objects like galaxies. In one of Tyson’s experiments, the Hubble Space Telescope was trained on a cluster of several hundred galaxies some 2 billion light-years from Earth in the constellation Pisces that seemed a good bet for a gravitational lens. Sure enough, Tyson picked up at least eight images or partial images of another galaxy “behind” the cluster, a systematic distortion that revealed the presence of a good deal of dark matter. Aided by the fact that individual galaxies inside the cluster served as smaller lenses, revealing fine details of their masses, Tyson and collaborators Greg Kochanski and Ian Dell’Antonio created a map showing the distribution of cosmic dark matter at unprecedented resolution. Their map was published this May in Astrophysical Journal Letters, with more data to come from Hubble and the special Big Throughput Camera built by Tyson and University of Michigan astronomer Gary Bernstein. Installed on a telescope in northern Chile, it offers 200 times Hubble’s field of view.
Tony Tyson might seem a throwback to the old ways, pursuing a fascination with no apparent relationship to Lucent’s business. But even he does not conform completely to the old Bell Labs model. While practicing his basic science, the astrophysicist has also worked on several applied projects. What’s more, while hunting cosmic dark matter, he pushed the development of charge-coupled devices for image detection and helped create novel image processing software-advances that have been incorporated into an automated fingerprint detection technology designed to replace locks, and a valuable failure analysis tool that maps the surface temperatures of semiconductors while they’re still in production.
Tyson’s work-like Alan Gelperin’s-can be taken to illustrate how Lucent’s attention to applications can pay off. Conversely, it can be used to show that companies should support unfettered science-because far-ranging studies have a way of paying dividends where they’re not always expected.
Indeed, the chief complaint from critics of the new Bell Labs is that the drive for relevance has overly constrained scientific inquiries-a strategy that will ultimately cause it to miss the kind of breakthroughs that brought the lab to glory. Many of the critics were drawn from the staff of the lab itself. Morale plummeted during the early 1990s, as the changes were implemented. Scores of veteran researchers quit; so many landed jobs at the University of California, Santa Barbara, that folks back in Murray Hill began calling the school Bell Labs West.
Former Bell Labs researcher Charles Townes, the Nobel laureate inventor of the maser and one of Arno Penzias’ instructors at Columbia, understands the reason behind the changes and doesn’t know what could have been done differently. Yet he feels that a good deal of Bell’s pioneering spirit is evaporating.
The loss is especially lamentable, he says, because more than almost any university, the labs brought world-class scientists together with experts in areas such as electronics or antenna design-producing a tremendous climate of discovery. “Bell Labs was a rather unusual and exceptional place,” notes Townes. “For a long time it could be different from other companies because it was a monopoly.” Now that it functions like any other company, he adds, “I think it’s a great loss for the country.”
While agreeing generally with Townes, Tyson says the dynamic for discovery may actually be better now than at any time since the 1950s. An increased focus on relevance has put short-term pressures on researchers and made it harder to pursue “pure” science. However, he states, “I think it’s healthy to have this tension. Otherwise you’re just sitting in the Ivory Tower doing nothing for anybody. It really does help to be immersed in the needs of the corporation at the same time you’re trying to make some new discovery. If you’re immersed in other cross streams of technology, of ideas, of demands…that’s a very rich environment for completely new ideas to spring forward.”
A third perspective comes from Penzias. He agrees with his former mentor Townes that some of Bell’s special qualities have disappeared. “There is a lot in what Charlie says, especially in the physical sciences,” he admits. “I have to say something has been lost. But that loss is not unique to industrial research. Nothing is what it used to be.” Especially not the reborn Bell Labs.
The big new idea for making self-driving cars that can go anywhere
The mainstream approach to driverless cars is slow and difficult. These startups think going all-in on AI will get there faster.
Inside Charm Industrial’s big bet on corn stalks for carbon removal
The startup used plant matter and bio-oil to sequester thousands of tons of carbon. The question now is how reliable, scalable, and economical this approach will prove.
The dark secret behind those cute AI-generated animal images
Google Brain has revealed its own image-making AI, called Imagen. But don't expect to see anything that isn't wholesome.
The hype around DeepMind’s new AI model misses what’s actually cool about it
Some worry that the chatter about these tools is doing the whole field a disservice.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.